An Improved Bearing Fault Diagnosis Method using One-Dimensional CNN and LSTM

Abstract

As one of the most critical components in rotating machinery, bearing fault diagnosis has attracted many researchers' attention. The traditional methods for bearing fault diagnosis normally requires three steps, including data pre-processing, feature extraction and pattern classification, which require much expertise and experience. This paper takes advantage of deep learning algorithms and proposes an improved bearing fault diagnosis method based on a convolutional neural network (CNN) and a long-short-term memory (LSTM) recurrent neural network whose input is the raw sampling signal without any pre-processing or traditional feature extraction. The CNN is frequently used in image classification as it could extract features automatically from high-dimensional data, while LSTM is most applied in speech recognition as it considers time coherence. This paper combined one-dimensional CNN and LSTM into one unified structure by using the CNN's output as input to the LSTM to identify the bearing fault types. First, a part of raw bearing signal data is used as the training dataset in the model, and the simulation ends when the number of iterations reaches a specific value. Second, the rest of the signal data was input in the trained model as the testing dataset to verify the effectiveness of the proposed method. The results show that the average accuracy rate in the testing dataset of this proposed method reaches more than 99 %, which outperforms other algorithms for bearing fault diagnosis.